Amazon Alexa is on pace to lose $10B this year
arstechnica.com
arstechnica.com
I wonder how much the decline in quality of Amazon's marketplace has affected other parts of the company's business including Alexa. I remember a time when, if I wanted to buy something online, I'd just buy off Amazon without shopping around. They almost always had the best price, their shipping was fast, and I trusted the quality of what I received. In that world, I could see myself saying my shopping list out lout to an Echo to make things easier.
But in Amazon's current free-for-all marketplace, I would never do that. Now I typically check traditional big box stores first, and only go with Amazon if I can't find what I need elsewhere.
Based on other search systems I have worked with, they probably have an initial fast filter that limits the items from the entire inventory to a loosely matching set (erring on the side of including irrelevant items). Then a more intelligent and expensive ranking system would be used to find the best items from that set (where “best” means a bunch of things including relevance and profitability). Selecting “price: low to high” probably bypasses the intelligent system, so you end up with crap at the top of the list.
The low-priced items could be doing some kind of SEO scam to get on the list, or it could be something innocent like a fuzzy match to text in the description (could be a dumb issue like tf-idf that randomly exceeds a threshold due to noise).
You could argue that if they don’t prioritize the investment to fix these issues, it’s in some sense intentional. Nobody sorts by low price, because it gives poor results, because they don’t bother fixing it, because it’s not a priority, because nobody sorts by low price…
This is the reason for this problem. When searching for 'cat' and sorting by lowest price, the first promoted item doesn't contain the word cat and neither does the first actual result (but they are cat items).
If they didn't hoist 'legitimately hard' problems onto themselves, maybe they wouldn't need to solve them.
I did a search recently for a specific part. "39SF010". (a ~USD1.50-3 flash-ROM chip that is backordered til next year from "legitimate" vendors and from the manufacturer themselves). This is one keyword, and a pretty damn distinctive one. There's little room for there to be a "helpful" algorithm to fix my typing into something better.
With the default results, the first page contained 48 items. Exactly ONE was a relevant product. As a customer, I'd have been perfectly happy to see "results 1-1 of 1" and called it a day. But Amazon clearly didn't want to admit their warehouses are not infinite. There would be some potentially competent fuzzy match options there, but they whiffed it hilariously. Show me some other flash-ROMs, or at least other ICs, programmers, or breadboards, or those starter-kit boxes of no-brand resistors and capacitors... but they instead listed a bunch of toothbrush heads, a box of cookies, three random power bricks, a barbecue grille, flourescent bulbs, and a set of power tools running north of $1500.
If you search for "samsung smartphone" and sort from low to high, all you get are covers and screen protectors in the first many pages.
I just decided not to buy an item from Amazon despite it being cheaper, because I need it and I don’t have time to deal with it being a model from a previous year or whatever the catch may be.
Hell, I don't even know how to buy anything via Alexa, and the one time I tried "Order some cheerios", it never came.
Alexa doesn't even push me to order things. It does, however, push me to use Amazon Music, which is a total nonstarter for probably everyone.
How did you come to this conclusion?
on HN.
I know many many many people that don't give one iota about the things you and I might care about regarding Amazon's "quality". All they care about is that they thing they want is available, fast/free shipping, and at a price they are okay with paying. They don't know one thing about comingling inventory, they don't care about fake reviews, or any of the other things discussed frequently here.
Really? Because Amazon is on pace to lose $10B this year
But browsing, yeah that shit is a nightmare.
It’s an open to all market, and there are certified channels and customers reviews. Big box stores dream to replicate that.
I still like it, as sometimes you don't really care if you get some cheap knock-off good.
But for important stuff like my monitor, television etc. there's no way I would buy them from Amazon.
How is it certified if they are commingling inventory?
So it presumably has to be hardware costs, but Alexa is software, and they have vertical integration of all servers which presumably would drive voice processing costs to negligible levels as well (and that's assuming 0 on-device caching/learning ability). And this is all assuming that the gross revenue generated by Alexa is $0, which also certainly isn't true. So I don't see where the numbers are coming from?
Also, you need to count the server costs, including machine learning training and inference.
Sell enough of these 30 dollar gadgets for a dollar and it will add up quickly.
Creative journalism or creative accounting.
That was what I thought as well. Whether to sugarcoat propr or upcoming layoffs, this niewspiece sounds like a justification in the public's eye of why everyone at Amazon is getting the axe.
(entirely speculation)
Consider it this way: if Alexa doesn't buy that capacity, someone else could, so it's important to capture this opportunity cost.
It's not perfect, however. I've seen examples where the repairs division of a company had to buy parts from its distribution branch at retail price. All well and good, so far. Except they then had to mark-up that part's price in the repair cost. Combine that with the fact distribution would give discounted rates to other repair companies to secure business and what you have is a repair division that's being outbid on price by competitors using the same parts!
I remember when I toured a large local forklift company (Brand you’ve probably heard of - big plant) and on the main production floor they had just about every OTHER brand of forklift.
Someone asked and was told they had to pay full retail (including all taxes) on their own units, and they have to buy their competitors products for research… so, two birds one stone…
Source: work at AWS
Think bigger picture. Imagine a panopticon spanning Ring.com camera networks and voice recordings telling you what's going on inside, with AI providing automated summaries of recent conversations. All delivered end-to-end by Amazon!
I’m more worried about smart phones acclimating people go carrying tracking and microphones. But, we find those tools indispensable now, so instead, we seem to want to focus on smart speakers
I find it hard to believe Amazon is spending more than $2 on payment and delivery.
https://asia.nikkei.com/Business/Technology/Amazon-Echo-tear...
Annual sales ~100 million device, then losing $50/device would lead to a figure of $5 billion loss in a year. Loss of $50/device would include cost for running R&D, running factories, giving minimum guarantee to factories, and all other associated cost - marketing, customer acquisition, support etc..
So $10B loss in a year sounds in the realm of possibility.
For example: I own a multiroom speaker setup not made by Amazon that is Alexa-enabled, all speakers could run Alexa if I had set them up for it (which I won't, ever), and they're very likely counted on this metric. I can remember a multitude of other speakers and devices with an Alexa-enabled badge on their packaging that aren't made by Amazon.
It'd be good to have hardware figures for sales of Amazon's own devices.
However, supporting all the 'partner devices' comes with associated cost in terms of engg' development and support (somewhat like what Android must be dealing with. Android division in Google would not be a low cost division I'm sure).
I was at MS when Windows was a 'thing'. Windows was purely about software, the complexity and support requirements scales in logarithmic scale as more and more OEMs gets involved with their unique hardware, business constraints and cultures.
What issues are you experiencing? Wonder if it’s something I can keep a look out for.
I dunno, it's the sort of thing that would be fit for a robocop-like dystopia just a couple decades ago but is seemingly... "normal" now.
If Amazon wants to datamine the dickens out of that to sell me more socks and USB cables, I really don't care.
Boggles the mind.
One internal document described the business model by saying, "We want to make money when people use our devices, not when they buy our devices."
That plan never really materialized, though. It's not like Alexa plays ad breaks after you use it, so the hope was that people would buy things on Amazon via their voice. Not many people want to trust an AI with spending their money or buying an item without seeing a picture or reading reviews. The report says that by year four of the Alexa experiment, "Alexa was getting a billion interactions a week, but most of those conversations were trivial commands to play music or ask about the weather." Those questions aren't monetizable.
Which doesn't include the overheads (engineering and staff, development and maintenance of back-end services, legal and compliance and all the rest) that you would normally plan to cover with the gross profit on the device.
Amazon seems to be a very poorly run company.
So many issues with search, with spam, with listings, and here's another data point...
Just bought a laptop. Waited until it was on sale, after doing Linux compatible research.
It shows up. 3 days later, it is even more on sale. This is a product sold and shipped by amazon.com, in both cases.
I contact support, and ask for a price drop refund, as there is a $150 price difference now. Nope, guy keeps linking me to "Price match" docs. I explain, I am not comparing vs another supplier, but amazon itself.
I am told, just return the laptop and order it again. That there is no other option.
For me, there is no return cost, other than a 5 minute detour to drop it off. Free return shipping.
For them, they lose the original shipping cost, the return cost, the processing of the return, and will sell it via amazon warehouse, for a $250 price drop.
I imagine that to sell a laptop via amazon warehouse, a check that it boots, a drive wipe is done, etc.
This is indicative of a company incapable of handling scale correctly.
Yet they blather on about loss on returns! And about return scams!
And yet, I still get the laptop at the price I want.
Only an insane person would act this way, and this insane act is a weird emegent behaviour at amazon.
It's telling, to me.
Amazon === borked
My guess is that they did some A/B testing and found that the "irrational" behavior actually saves them money. In your example, if 50% of people don't bother going through the return process they still come out ahead.
For example, you consistently buy small ticket household items and rarely return anything: when you do return something they will be helpful and generous to keep you happy.
You buy large amounts of fairly expensive clothes and shoes but return around 50% after trying on. They will accept the returns without issues but won't let you keep anything.
You're a bargain hunter who buys expensive laptops or hard drives whenever the discounts look really good. They will make returning stuff a little harder for you since they know you don't have loyalty, and they don't want to facilitate arbitrage.
Doing this is computationally no more sophisticated than the work they are already doing to detect professional scammers. They aren't supposed to show different customers different prices, but they can freely use things like this to do stealth price differentiation, since many details of how the return policy is implemented are at their discretion.
I've always assumed there's some sort of credit score system built into every retailer account, and am somewhat conscious to try to keep it positive.
Insanity. It’s tripping over dollars to save pennies.
Although you returned the laptop, the annoying process might have discouraged many others from doing so. The shipping cost and the cost of processing the return are likely much smaller than you expect when accounted for on Amazon's scale. And by making you return and re-order, they are putting the risk on you if the laptop is not the identical product, rather than taking on the costs of administering a price drop refund themselves.
Of all the things that Amazon do, analysing the game theory of customer behavior so that they can do as well as possible out of interactions like this seems like one of their biggest strengths. They certainly have thousands of skilled people working on it.
As for return scams, I would not take what they say at face value. My opinion is that they have quite sophisticated models for analysing returns, both for detection (to distinguish between the behavior professional scammers and regular customers) and economically (to decide that it's cheaper to give the latter the benefit of the doubt, let them keep returned items, etc). The only flaw is if otherwise legitimate customers start to think they can get away with falsely returning 1 in 20 items or whatever. So they 'blather on' about return scams, not because they are material to their business, but to make it clear that if you cross an invisible line, they will go from not caring about your $120 air fryer that you ended up with two of, to pursuing serious legal and criminal consequences.
I’d say that Amazon has the data to determine if it’s more profitable to give the customer a credit or have them ship and return it.
So while annoying, I’d say that scale is probably why they are doing this seemingly unintuitive thing.
But if you were accounting for the profit or loss of Alexa specifically, of course you would account for all the AWS services used at the price they are sold for externally.
Having spent tens of billions on something that doesn't add up to the bottom line doesn't go well with shareholders, especially when there's no prospect that it ever will.
Alexa, time! Alexa, set an alarm for 5 minutes! Alexa how many minutes left? Alexa, turn off/on all the lights!
That’s basically it. I wrote some apps for it a long time ago to do custom stuff like read me some Reddit pages, but the SDK changed or something and eventually they just died and it wasn’t obvious to me how to recreate it/not worth the effort. I really wish there was an easy way to just put python scripts onto the device or something. The process of going through Amazon is pretty unnecessarily complex and annoying.
I know there are ways of doing this stuff with raspberry pi, but also: not really worth the effort. If I cared that much I’d just make a PWA for my house and give that to my wife and kids.
Doesn't https://mycroft.ai fit this purpose?
Looks like you can ssh in: https://mycroft-ai.gitbook.io/mark-ii/advanced/ssh-access
and has python running on it:
https://mycroft-ai.gitbook.io/docs/skill-development/introdu...
Has anyone got one though?
It seems to be $350 (no international pricing?) which seems far too high to be realistic. More subjectively, that's quite an ugly-looking unit.
Also, not shipping till Jan? I'd be interested to see a review if anyone has managed to get one.
The dev teams are so damn focused on being able to answer some esoteric query that you're asking as joke, they forget that you only want to do like 5 basic things:
* Set an alarm/reminder
* Read then send a text message (while driving). It doesn't even have to be good. Just something that I can send while driving.
* Maybe.....maybe ask for the weather.
* Turn a named household device on/off.
* Play a song
It's like these services all completely overlook the actions that you'd be taking on your phone anyways.
But I also recognize that my usage won't generate revenue and will continue to cost. Surely they can't just keep offering free service forever on equipment that was purchased at deep discount. But I don't think offering gimmicks is the answer to this.
I remember one time I asked my Google Home the weather to see if it was cold enough to put on a jacket, and after hearing a temperature that I was ambivalent about, I took a risk and tried asking it what the wind speed was. It started reading out the Wikipedia page for "wind speed" until I told it to stop.
Worked for me
A lot of apps have tie-ins to Siri but it's very hard to discover them and understand how to use them. I think the negotiation of the shared language between us and our robot servants/friends/overlords is going to be difficult until we can answer that question.
Sometimes I want to listen to a podcast without risking accidentally spending the next hour on Reddit.
This betrays a spectacular failure of imagination on improving what the system can do. Show me a weather map. Show me turn by turn directions. Show me the television schedule. These are simple things that the display version cannot do.
Another problem is the clunkiness of accessing the apps. Ask foobar to tell me a joke … why not just “tell me a joke,” and remember that foobar is the preferred app to handle it? I have to remember two things. I have to remember that foobar is the app that I want, and tell me a joke is the command I must give to it.
Finally, volume control. When I ask it to turn it up, it stupidly increments it by one notch. Instead, it should take a sample of ambient noise, and pick a setting that is slightly above that. Otherwise, why would I be asking it to turn it up?
Just really simple simple things. I bet they could take 10 very smart people, that have never used this in the first place, and just watch them while they imagine all the possible things that could help around the house.
And then one time out of 10 or something it added crap like "by the way, if you ever wanna know about blabla, you can just ask me to do blabla".
Now I don't own a google home mini anymore.
Then, the only integration you need is Alexa-> Home Assistant, so there is only one thing that can break.
- To ask for the weather, weather alerts, relative humidity, precipitation, wind speed
- To ask about outdoor air quality, indoor air quality (requires air quality monitors)
- As a baby monitor (crying detector routine)
- To know when to empty the dehumidifier bucket (humidity > 60% routine)
- To know when someone left the AC fan off (high CO2 routine)
- To know when to open the window (high VOC routine)
- To know when to turn on the HEPA filters on high (high PM routine)
- To turn lights or on off
- To manage the thermostat temperature
- For music during lunch/dinner
- Reminder for meetings, taking out the trash, etc.
- Timers, alerts, alarms, reminders...
- To automatically turn off devices at night via smart plugs
- Vacuuming, mopping
- To locate my phone by calling myself
- Stock prices
I couldn't live without it.
Maybe not the most lucrative revenue stream, but not nothing.
Please elaborate.
I would expect that in the future, home automation does all the regulation for you without any interaction. You set up a bunch of configuration in advance, and maybe train the device a little bit initially on things like your temperature preferences (because we think we want a certain temp, but actually usually find it slightly too warm or whatever). Then you almost never have to make changes, let alone expend your executive function on deciding what instructions to give the devices.
It's horrifying to me that anyone would allow such a device in their home. Someone gave me an Echo as a gift once. I smashed it and buried it in the back yard.
But what they forget that the only reason movies and TV have it in the first place is because that way the audience hears what the character wants!
It's not because it's actually a better interface - because it's not!
Same with glass interfaces with no buttons - they make them because they are cheaper, not because it's better. But they look really cool, so people start to think they are actually better.
I'm sure there are tons more examples of things people think are done because they are better, but are actually done to save money or for some unrelated reason.
Maybe shoelaces that are laced inside-out. In the box they want the ends to point inward, but when you wear them you are supposed to relace them so they end up pointing outward.
> Same with glass interfaces with no buttons - they make them because they are cheaper, not because it's better.
But are you saying current smartphones are somehow worse than button phones? I'll pass on that. And they ain't cheaper too :)
That's also why Siri is good enough even though it is destroyed in a trivia contest against Google or Amazon.
Voice UI is very bad since you don't have a way to know what it can do reliably. Treating it as general artificial intelligence doesn't work because the experience feels like trying to get a really stupid person do something for you. It's much easier to do it yourself than to find a sentence structure that works for this particular request but not for others.
Voice UI is only good for basic commands that you know the device is capable of.
>> I'm sorry I can't do that
> Ok Google clear notifications
>> Opens Bluetooth (!)
> Ok Google WhatsApp call xyz
>> I can't call with duo (!) yet
> Ok Google restart phone
>> To restart your phone <instructions>
For first few years, Kindle Reader was a massive sink hole at multiple levels. There was this humongous R&D cost, hardware and software. On a recurring basis Amazon absorbed cellular data charges. Remember this was in early 2008 so cellular data charges were big and since this was the first couple of iterations of product the data sync protocol was yet to be optimised so data intensive. For example, the device would phone home every hour or so to fetch new book purchases and metadata such as bookmarks, last read pages etc., It also included a basic web browser. And finally, the unit economics of selling digital books were terrible. Because Amazon was trying to create a new habit they had to incentivise readers with discounts, even without discount they were losing about 4$/book sold.
So all in all this all seemed like a giant resource consuming operation with no hope of generating any profit. But boy did it succeed. Within about 4 years or so kindle books were outselling physical copies something like 3-1. And now I think it's not even worth comparing.
My suspicion is execs expected Alexa ecosystem to follow Kindle's trajectory of success which doesn't seem to have materialised.
I think that sentence encapsulates the problem well: Kindle could succeed as a product with a very simple KPI (ramp up economies of scale until you make a profit per book sold), while Alexa has to develop some nebulous "ecosystem" that nobody is quite sure what it could possibly look like.
You just wanted to sell a new way towards it.
But audio-things aren't even properly defined.
Probably most of us who were kids before remote controls were common remember changing channels and turning off lights and opening garage doors, all in response to voice commands from our parents.
Amazon has fuelled the race to the bottom and it seems the company inevitably destroys its own good creations given just a short time.
Alexa pulled talent from everywhere throughout Amazon. It took up a huge portion of the hiring funnel for new devs. All of those people could have been doing useful, money-making, cost-cutting things instead.
I can think of at least a dozen really brilliant devs I worked with in Amazon Fulfillment who left to go work for Alexa. They'd tell me what their team/group/org did and I'd ask "but how does that make money?" and they'd laugh. "Yeah, last year I heard we had less than a million in revenue, but hey, they're certain it'll become big somehow".
What a waste of talent.
> “As a company grows, everything needs to scale, including the size of your failed experiments. If the size of your failures isn’t growing, you’re not going to be inventing at a size that can actually move the needle. Amazon will be experimenting at the right scale for a company of our size if we occasionally have multibillion-dollar failures.”
[1] https://www.aboutamazon.com/news/company-news/2018-letter-to...
Lots of services built for Alexa have been spun out into AWS as well.
It's really baffling how Amazon is actively getting worse to buy from, and sees no problem with it.
This is how you kill a tech company: stop innovating and fall back to simple “maximize revenue, cut expenses” MBA theory
Amazon's innovation age is dead, imho. They are the Day 2 company that Jeff Bezos always warned about.
The other challenge though is doing 100 other things half-hearted. Perhaps they should be MVPing smaller ideas to see if they take flight, but ambitious projects always require substantial resources.
The complaint “this could have been amazing but leadership simply didn’t fund it enough” is another common gripe.
There’s definitely some craziness in the idea that everyone will order toothpaste with their voice and it will work well and replace an application to the point it will be profitable.
What problems should a sophisticated home assistant solve? It should curate and offer SOLICITED services; find me washer dryer, fridge, plumber, landscaper, house/pet sitter, house cleaner, mechanic... It should present these to me visually to choose from. It should remember and be ready to automate any subsequent request for the same service. This could be done via integrating with reputable third-party providers like Angie's list, thumbtack, google flights, aggregators... Alternatively, Amazon should build reputable service provider network itself. People need goods and services. Provide a path to solicit, opt-in to targeted curated offerings/ads. As important, provide the ability to opt-out when no longer necessary. Unsolicited ads are often inaccurate, invasive and damaging noise. Earned reputation and trust is gold.
We've tried to use Alexa for some games, like Jeopardy or Whose Line Is It Anyway, but it always is an incredibly buggy experience. Even just simple multiple choice games should work perfectly, and yet, a lot of the time even basic answers don't register, and the games themselves crash.
I've always wondered - stupid mobile games are dominating video games in terms of money right now. Voice platforms seem like they could do something similar, but all the platforms want to be your shopping AI or whatever. Just seems like the two worlds - voice platform and game developers - just could never get together and figure things out.
1. cooking timers
2. fart noises
3. add X to my shopping list (used when shopping in a physical store)
4. Weather forecast
5. Turn off the tv
That's enough that I'm happy 'she' was built into our tv sound bar, but yeah, I don't think alexa is making any money from me.
I think that everyone on here, that is like you and me, is that way because we have trimmed down our expectations over time due to poor implementation.
When money isn't cheap anymore, and you have to start showing a profit to get more money, the screws turn and users get squeezed. Merely showing a profit doesn't mean anything if you can't afford to do R&D without funding. Or, for a manufacturing perspective, your factory turns a profit, but you need to borrow money to buy a million dollar machine. The more it costs to borrow that money, the worse it gets for your customers - either because you use the older machines and make inferior product, or you have to raise your prices so the loan doesn't eat in to your profits as much.
Netflix for example has a lot of costs — servers, payroll, content creation, licensing, etc.
Interest is one cost, but hardly the whole picture of them. How exactly does Netflix’s interest costs going up mean their business model fails? Does it mean they must charge $10.50 instead of $10 per user?
It's not the interest, it's getting the money at all. Money was cheap, so netflix got lots of money, they put blockbuster and hollywood video and all the other companies out of business (mostly) with a great selection of DVDs and whatnot, then they got sweetheart deals from all the studios and production companies who couldn't imagine competing on the bandwidth costs alone; these deals, once again, greased by investor money. Once the major studios started their own streaming services, netflix was throwing investor money at consumer eyeballs with their whole "sharing is caring" model, where you're aunt's ex-husband's golf buddy could use your account credentials.
Now the free money has dried up, and investors want, you know, their investments back. Queue netflix raising prices, disallowing sharing accounts (and charging more for multi-screens); couple that with losing a huge portion of their library to studio streaming services, and suddenly paying adam sandler 100mm for a movie every year or two looks like a bad investment.
And there is latency issue too, you won't get response very fast. If you look for <0.5s latency or the answer, you need to try something like Vosk probably.
Plus everyone's mobile device is already this anyhow.
People today: Hey wiretap, do you have a recipe for pancakes?[0]
https://techcrunch.com/2018/11/14/amazon-echo-recordings-jud...
They complied with a legal order - they didn't submit it to the FBI.
Do you need some more tinfoil for your hat?
Frankly I'm baffled at how you could be as confused as you claim to be.
My toddler listens to the animal of the day at least once a day on it, comprising >50% of my overall voice assistant usage, and I love to hope that I'm a weird outlier on some confused analyst's usage dashboard.
Fun fact, if you Google voice search animal sounds on a pixel, chrome will (most of the time) serve up a whole animal sounds dashboard with pixel art.
The toddlers love yelling at it and getting it to play songs they like. The older ones, well, I guess do the same thing but with different songs, and occasionally Wikipedia-style quick answers.
My gf has one in her living room. It is only used to play music, set a timer (usually watch is used) and adding things to a (third party) grocery list. Plus it's listening all the time to what goes on.
I had one alexa device: a scanner that would read a barcode and add it to a shopping list (or you could talk to it -- ugh, had to keep it in a drawer). It was handy as I ran out of milk to just scan it and forget about it, so that I would see "milk" when at the shop. But they discontinued that device.
"Echo devices use on-device keyword spotting designed to detect when a customer says the wake word. This technology inspects acoustic patterns in the room to detect when the wake word has been spoken using a short, on-device buffer that is continuously overwritten. This on-device buffer exists in temporary memory (RAM); audio is not recorded to any on-device storage. The device does not stream audio to the cloud until the wake word is detected or the action button on the device is pressed. If it does not detect the wake word or if the action button is not pressed, no audio is sent to the cloud."
https://d1.awsstatic.com/product-marketing/A4B/White%20Paper...
The speaker also starts recording a few seconds before a command is issued, meaning there’s likely more private information in the recordings than customers are aware of. from https://www.vox.com/the-goods/2018/11/12/18089090/amazon-ech...
https://www.nytimes.com/2016/12/28/business/amazon-echo-murd...
Those two links are from two different murder cases for which Alexa/Amazon data was subpoenaed. There are probably others.
Which assumption is the safer one?
I've waited my whole life for AI assistants. I remember being really excited about the promise in the 90s. But it's this is it? Voluntary advertisement consumption?
Very disappointing.
Alexa arriving with a default profile for generic recognition makes sense -- but after that, it should have learned me, my voice, my commands.
Instead, it was one more not so subtle demand from Silicon Valley to become synonymous with their Product Persona -- or be ignored.
I have a google smart speaker thing precisely because it's the only device that I can integrate with my Nest cameras and devices (which I bought before the Google acquisition). It does 2 things: 1) announce door bell ringing, and 2) play Apple Music in my office.
For everything else Siri, whilst demonstrably behind the curve of Amazon and Google, is perfectly functional - dictating short messages whilst walking, adding to a shared shopping list, setting timers.
I already have the phone for other reasons, there's no use case for buying an Alexa device that I can see.
I would never make a purchase from Amazon on an Alexa using my voice - there's simply too many variables that I can't control for. It might misunderstand (possibly catastrophically costing me a huge amount of money and inconvenience), the quality of goods on Amazon is too low and generally too variable to allow a an Amazon AI to be the arbiter of which product to buy for me, and even the manual purchase process changes too often and too subtly to have confidence that it won't add something that I don't want (i.e., a regular subscription to a product, pay on a credit arrangement, etc).
This hasn't worked for me since iOS 13, I think, and macOS, too. Getting a "Sorry, can't help with that".
"Alexa was getting a billion interactions a week, but most of those conversations were trivial commands to play music or ask about the weather. Those questions aren't monetizable."
This is mostly how I use Alexa, but those requests to play music are backed by an annual Amazon Music Unlimited subscription that I wouldn't bother with if I didn't have Alexas throughout the house.
Back when it was only Amazon it may have been possible to Alexa to buy some very specific thing (“buy Mario Party 23 for Switch”) but. Now I wouldn’t be able to trust even that.
So say nothing of worries about kids/guests/accidental activations ordering things.
But it never happened. Alexa never got smarter. It still misunderstands basic stuff. You still have to use specific words for specific actions or it gets confused. You still have to maintain silence for a couple of seconds before speaking to anybody else after a command otherwise it will go into a loop of "I don't know that". It still won't play the exact song I want to.
The development experience is also extremely limited and stunted. You cannot run any code on the device, period. You can only define lambda functions that are triggered by voice. The function does not have access to anything from the device, only the spoken words.
You want to incorporate the temperature of the room using the built in sensor? No can do.
You want to know the state of a bulb or a switch? Not possible.
You want to do things on a schedule? Nope, it has to always be in response to a command.
So I'm not surprised they are losing money. They dropped the ball hard. They built an amazing thing and let it stagnate. They've steadily improved the hardware, but totally ignored the software. And software is what's magical about this. A robotic person who talks to you! This has to be pure software play. Hardware is secondary.
The problem with voice assistants at the moment is the same as command line tools.
You have to know what specific prompts work and don’t work etc. I often see my parents make simple queries like “turn off the white lights and turn on the night lights” only to get the response that it can only do one thing at a time.
There will be a future where this gets more and more sophisticated. But in the meantime, it’s not surprising it’s losing money.
I agree completely with your assessment, that they behave more like a verbal CLI than a proper assistant, but solving that well doesn't seem like it offers a great opportunity for monetization.
My hope is that the open source voice assistants get more popular, and it gets easier to flash the Amazon Echo to be a satellite device for the open source voice assistant. I've looked at running the open source voice assistants locally, but the biggest barrier I had was the cost for each of the microphone/speakers. Amazon's really got the cost of those things way way down.
"Alexa, order two medium cheese pizzas from the closest Domino's for delivery."
"Okay, ordering two medium cheese pizzas from a Domino's 2.7 miles away for delivery as soon as possible, for $25. Is that correct?"
"Yes."
"Okay, ordering pizzas. It'll take around 40 minutes."
But you'd need a very high level of AI intelligence/integration to get to that level. As it stands you'd probably go back and forth a bunch of times as the AI quizzes you on exactly what you want, when you want it, from which restaurant, etc. And that would just kill the usability, to the point where it's better to just use your phone.
And then, there's a weird thing where this probably only works for the simplest (and then cheapest) transactions. If I'm placing a complex order because I've got 7 friends over, I'd rather use an online ordering system or just make a phone call.
So you're taking a small cut, and probably only getting in on the smaller orders. There'd need to be a lot of volume to make it work.
https://www.speechrecsolutions.com/mic_guide_handheld.html
Turns out they're third party devices. That makes Dragon more impressive than I thought.
But that's not a voice assistant, that's just voice transcription.
HomePods aren’t sold for a loss (Apple doesn’t do that, plus price makes it seem even less likely). Macs/iPhones/iPads/etc also pay for its development.
But it doesn’t exist to make money like Alexa (why would I buy random stuff with it?). It exists to make Apple devices more sticky. And it works well for what I use it for.
The Alexa business model never makes sense to me, and that’s even if the devices were never sold at a loss, which they frequently were.
Edit: cleanup
[1] https://www.nytimes.com/2022/11/14/technology/amazon-layoffs...
I have a Mycroft device that I'm playing with, but for the life of me, after years of using these devices, I cannot imagine getting a lot of utility in my home life.
I do think voice activation is great in certain circumstances, like when I'm driving, or riding my bike, but the rest of it does not offer much to me.
The market penetration is therefore finite, limited to geeks and early adopters. I'm somewhat surprised they didn't consider profitability more thoroughly when they started with Alexa. It must have cost them $100 billion by now.
"Something that makes my life easier! Cool!"
People don't see any value in it. And some technically inclined acquaintance will probably point out that the thing records everything they say and sends it to Amazon.
The only thing that mildly makes sense is something like an offline version that controls the lights, heating etc.
Search is not Amazon's or Apple's main business so they will have a hard time competing. (last I checked, Siri just does a Google search and provides links when it doesn't understand a question)
However, both Amazon and Google seem to sell their devices at cost with the hope of finding a revenue model while Apple charges a premium for their Home Pod. That makes the Apple approach most sustainable at this point.
It’s really annoying.
1. Set timer
2. Convert weird US fantasy units to sane ones
3. Play music
4. Control lights
5. Weather
6. No, I don’t fucking care about the notification of some shipment I already got on my phone and my mail.
Alexa devices are cheap. Seriously cheap. That is the main reason I have not replaced them with Rhasspy [0] yet.
A decade and 25,000 Alexa "apps" later and that's still just about all that anyone does with it.
The Alexa team are (were?) absolute zealots about "Voice UI" and insisted that the entire world was going to be run by voice. You couldn't convince them that voice interfaces weren't the perfect solution to every problem. They were under this delusion that Alexa would become so intelligent it could just figure out what you wanted without you needing to specify - as if HAL level general AI was just around the corner. Every user interface would become voice based. Absolute zealotry.
Combine this with the Rube Goldberg-esque back end where for years literally any "speechlet" implementation had the chance of bringing down the entire service and it's unsurprising this is the result.
I was fired a year or so after I came up with a method for interacting with Fire TVs and Echo Shows which solved the "voice cliff" problem by cutting out the round trip to Alexa and wasn't based on magical thinking. My product plan that I had prototyped and worked on for nearly a year was rejected wholesale.
Obviously, I'm extremely bitter and biased, but I swear to you it was that sort of stubborn cult-like thinking which lead to this headline.
Voice isn't the thing, it's a mode of interacting with the thing.
The thing is: the digital home.
Amazon is comically dropping the ball here. They have the tablets. They have the TVs. They have the reach. They have the home security division. Integrate, integrate, integrate.
Alexa is just a mode of communicating with the digital home. It's a feature of the digital home, not a monster profit product by itself.
It's like thinking a stylus is where you'll make money on a business tablet. No, it's a way to input.
I agree with your point on integration. All the pieces seem to be there already. I would not want it, but I clearly was not target audience to begin with ( and this stuff was flying off the shelf life crazy ).
Proton/OS/deck all just seem like hedges against their constant enemy: Microsoft. That was always the existential threat to Steam, and MS made it official when they launched their store.
It’s great it worked out, but I don’t think what we have today was in any of the early plans. I think most of it came later.
But that’s just my guess.
Sometimes companies don’t even have that and you get spastic, impulsive, uncoordinated, un-driven, near-net-zero movement, like a defibrillating heart.
I also created a free skill for kids to hear the daily school lunch menu. Such a morning time saver! https://github.com/jeffsheets/alexa-papio-lunch-menu
But yeah it’s still mostly a hardware solution looking for a problem
I am imagining GPT-4 on the other end with a suite of controls. If Alexa could understand what I'm talking about and all of the options available to it I think it would be a much better system. I could also imagine a system that's always listening and proactively acting on my behalf without my even giving directions. If hardware advances to the point where I could run a model this size locally? That would solve privacy concerns and improve latency. How far are we from running GPT-4 on commodity hardware? 20 years?
Siri has always tried to be more conversational (though not by too much). But that’s also probably one of the reasons it hasn’t been as expandable.
Much easier to add skills (in Amazon parlance) when the command can be “tell Foo to bar the baz” instead of trying to intuit which app you want to use based on heuristics or something.
I also heavily use a routine that reads out the weather forecast, upcoming appointments and the latest news.
I've def been a little annoyed at it starting to ask me for product reviews and other things these days, but most of it's been low enough friction that I don't bother shutting it down on the app or whatever.
I wish Alexa was more programmable or less clunky for a casual enthusiast though, I set up a buncha things basically the first few years and slowly had it atrophy as other priorities in life took over...
I was also struck yesterday by news that Disney loses a run rate of $6B annually on streaming with nearly 250M customers.
So they are losing ~$24/customer per year, selling a service at $8/month. It's interesting how all the competing streaming services popped up around $8/month that forced Netflix to offer lower priced tiers from their $12/15/20 range, but arguably that was actually the natural, profit-making price.
It's not a customer acquisition cost if you continue to lose it every month.
Anecdotally I can give my frequent Alexa commands for Echo here:
- how’s the weather today
- what time is it now
- turn on the TV
- set an alarm for 6am
- play 88.5 fm
- what’s Arsenal score right now.
Now it’s hard to understand how these kind of interactions can earn any money for Alexa.
There's probably a market for advertisers who want their victi- ahem, audience to hear their ad first thing in the morning.
I was really excited about Alexa at first, but when I tried it on a day to day basis the biggest issue I had was "Amazon knows too little about me, compared to Google". I know right ! How horrible is that.
But consider, Alexa didn't know what i like to listen to music wise, didn't know my favourite tech topics or news websites. It had almost zero history on, even though I have bought probably about 20 books in the last 10 years and from Amazon.
Amazon Alexa is a “colossal failure,” on pace to lose $10B this year https://news.ycombinator.com/item?id=33700160
Amazon didn't want the first route so they could gain market share/penetration quickly, so we're stuck with options 2 and 3. Voice assistants are a new media and people hate ads, especially where they are not expecting them - so no wonder they failed to insert ads there. We're left with what I believe they'll end up doing - requiring a prime subscription.
I don't trust it in the slightest.
I don't feel inspired by the awesome potential to allow me to creative and fun things.
That's why its failing.
And if it's "server costs for processing requests", then how come Google / Apple aren't losing $Bs on their VA systems? (Also, if no one is using Alexa then request processing costs would be low; if everyone is using Alexa then it seems there would be some return in higher sales.)
My kids use our Echo Dot to play music using our Amazon Prime subscription all the time. It's what makes me renew that subscription when it comes due.
But I bet most people don’t. They probably use them for competing services like Spotify that not only don’t make Amazon money but reinforce the feeling of not caring about Amazon music.
I’m in the Siri world, but I’m with the other posts in this thread. 99% of my usage is simple stuff like timers, math problems, smart home, etc. stuff that just won’t ever drive anyone’s revenue directly.
I've wondered if Alphabet's X/Waymo division operates in a similar manner. Giving future Alphabet a way to cut costs when it's growth stalls.
With Apple how long can the worlds biggest (or close, don’t know ATM) company keep growing?
So they’re doing user-hostile stuff like more ads and unnecessary subscriptions.
How many nice SaaS things had a perfect product bad had to pivot to enterprise just to find growth, ruining it in the process?
The popular perception that Amazon rakes in massive profits is some sort of folk wisdom.
Please write it off, free the patents and let people who actually care about it take the helm.
Drop the money into home assistant and let them write special integrations just for you. Be -the company- invested in open source
Well... Amazon would want you to do something with it that makes them money.
They also own a “platform” in the way that Zuck is trying to do with Metaverse at much greater cost.
All things considered, I don’t think it’s been such a bad investment and I don’t see it being killed.
Amazon is familiar with employing that strategy, and it hopes to make it pay off again.
But I am surprised it’s losing it this fast.
Amazon is nixing CloudCam, which rendered my Kwikset Amazon Key front door lock relatively useless.
If it is all loss making or tiny low single digit profit margin, then a $10B loss is still a $10B loss.
Several of its divisions are making a ton of money and even retail could be making more, but it's constantly reinvesting to grow + it's a loss leader.
Up to a point, I can admire the product discipline that Amazon and Apple (Alexa and Siri) have shown in keeping their voice products focused on consumers. But after so many years, when do we finally get voice for the enterprise? At one time, I was very hopeful about this space, see: "I believe in Enterprise software for the Amazon Echo"
http://www.smashcompany.com/business/i-believe-in-enterprise...
In late 2015 and early 2016 this space, of voice tools, was moving quickly and I thought that surely, within a year, we would see better support for enterprise applications. I was mistaken.
Here is an example of the excitement of those times: "Invoxia will allow Alexa to figure out who is speaking"
http://www.smashcompany.com/technology/invoxia-will-allow-al...
At that time, it felt like we were on the cusp of using voice tools in business. There was certainly a market. For instance, https://www.jwt.com/history/ has about 9,000 employees in at least a dozen countries. They reached out to me to talk about their "Pangea" project, which was a huge internal project to better map what teams and skills and resources they had internally. I had a meeting with them in the summer of 2016 and they asked "We are thinking we could put an Amazon Alexa in every meeting room, and then whenever we need someone with a specific skill, we could ask, hey, Alexa, do we have someone at this company who has this skill?"
At that time, I told them of some of the difficulties, but I was also hopeful that things would change a lot of over the next year. I was wrong.
Among the many, many problems we faced was simply getting Alexa to recognize the names of companies, when the sound of the name did not match a standard English spelling:
TATCHA
L’Oréal
L’OCCITANE
What was needed was a way train specific conversions of phonemes to text, for both Alexa and Siri. There are a small number of tricks that are available, such as programming a glottal stop:
http://www.smashcompany.com/technology/using-a-glottal-stop-...
However, in general, neither Amazon nor Apple wants to give us the tools we need, and there has been little progress in recent years.
I know that many people on Hacker News have read my book "How To Destroy A Tech Startup In Three Easy Steps" in which we were using NLP technologies to try to allow salespeople to send text messages to Salesforce:
https://www.amazon.com/dp/0998997617?psc=1&ref=ppx_yo2ov_dt_...
As you can imagine, we hoped to move on to voice-to-text so that salespeople could simply talk to Salesforce.
It seems to me there is a wave of new startups that are waiting to move forward in this area, but no small startup has the resources to build a voice-to-text system from scratch. We need the larger firms to help set up the ecosystem, so that we can move forward in this area.
Where is the company that is going to give us the tools we need to start building great voice driven enterprise solutions?
Most of the voice solutions I've seen kind of nibble around the edges of this, don't get traction, and slowly die a slow death somewhere deep in the middle of the RFP or evaluation process.
-------------------
I asked him to run me through the pitch, and John gave a practiced recital:
Most salespeople are human-centered and enjoy talking with other individuals — but they hate dealing with computers. If a salesperson is selling shampoo to Sheraton Hotels, the best part of their day will be talking to the customer; the worst part will be when they have to go back to the office and deal with their company's reporting software. More likely than not, this will be Salesforce, the most widely used software for tracking sales.
Salesforce is ugly. Their interface is clunky. The poor salesperson has to sit down, bring up the website, click on a bunch of buttons, and navigate through a bunch of forms. The worst day of high school math was probably more fun for them.
Celelot aimed to change that. Instead of dealing with Salesforce, the salesperson would simply pull out their phone and send a text message to the Celelot system. For example, "Spoke to Carol. I just sold 1 million bottles of shampoo to Sheraton Hotels, rev 500,000. Contract August 1. Delivery September 1." We would use a set of computer techniques known as Natural Language Processing, or NLP, to take a message like that and pull out all the fields that were significant to Salesforce:
Contact: Carol Harrington
Customer: Sheraton Hotels
Product: Shampoo
Quantity: 1,000,000
Revenue: $500,000
Close Date: August 1
Delivery Date: September 1
Celelot would automatically identify who sent the message, connect it with their Salesforce account, and log the information in the system. Salespeople would never have to interact with Salesforce directly.
Apart from streamlining the reporting process for Salesforce specifically, Celelot could potentially become the default interface for all sales-reporting software (a category officially known as Customer Relationship Managers, or CRMs, of which Salesforce and Pipedrive are two well-known examples). That would be game changing.
But the real question to determine if it's a viable product/business is does it drive enough value that people will actually buy it, use it, and remain paying customers?
I'd say in your example above, it likely won't massively decrease costs (the salespeople are still having to interact with a CRM) or drive increased incremental revenue in any real way.
Even my ecobee thermostat does not require an internet connection.
https://support.ecobee.com/s/articles/Wi-Fi-FAQ-Setup-Guide-...
> The short answer is yes.
>Even without a Wi-Fi connection, your ecobee will still function as a traditional thermostat; it will engage your equipment as needed and maintain your comfort set points.
There's two sorts of it, as I see it, and both of them do have weak spots:
* Turning everything into a data vaccuum to sell to marketers. Something like the GDPR but angrier could pop that balloon pretty decisively, as could shifts in marketing trends if advertisers start considering tracking critically as a cost centre, rather than the current FOMO snowball.
* Unbundle bits and pieces of the experience to sell back to you on a subscription. This still feels like the data's not in yet for a lot of products.
I suspect we see it being trialed a lot with cars because so many of them are leased and therefore have a predictable trade-in life. As long as the annoying $10-per-month service comes with a three-year trial, the original lessee never needs to think about the subscription model, it more becomes an albatross to make the used car look less appealing by comparison. It's harder to pull that particular trick for other products, as their service lives vary widely. I suspect that, for example, many home appliances with a subscription would end up with a lot of angry and confused calls to support when the machine "stops working" at the end of a trial/missed payment/etc. Having to pay for that support burden and associated brand damage may unwind most of the profit the subscription yielded.